workflow-mcp

Turn the AI workflow templates in the workflow-templates repository into MCP-callable tools — any MCP client (WorkBuddy, Claude Desktop, Cursor, etc.) can directly list templates, inspect contracts, execute templates, and get structured results.

Architect of digital pipelines · Fourth building block: make all pipelines "plug-and-play"

What it can do

Example session:

1. list_templates                    → 发现 financial-report-analysis
2. get_template_info(模板id)          → 拿到输入契约（需要 company_name / report_path / report_type）
3. run_template_async(模板id, json)   → {task_id, status: running}
4. get_task_status(task_id)           → 轮询至 done，返回指标 + 风险信号 + 报告

Why the async version is needed: clients like WorkBuddy / Claude Desktop impose request timeout limits on individual tool calls, while real template execution (multiple LLM calls) usually takes 1-2 minutes, so synchronous calls time out. The async version avoids this limitation.

Related MCP server: MCP Boilerplate

Architecture

MCP 客户端（WorkBuddy / Claude Desktop / ...）
        │  stdio / streamable-http
┌───────▼───────────────┐
│  server.py (FastMCP)  │   list / get / run 三个工具
└───────┬───────────────┘
┌───────▼───────────────┐
│  templates_registry.py │  扫描 workflow-templates，读取契约元数据
│  runner.py             │  子进程执行模板 run.py（--auto-review 自动化复核）
└───────┬───────────────┘
        │
┌───────▼───────────────┐
│  workflow-templates    │  模板仓库：契约 + 提示词 + 脚本（数据权威）
└───────────────────────┘

Key design decisions:

• The template repository remains the single source of truth: adding or modifying templates only requires changing workflow-templates; this service needs zero changes

• Automated review: run.py adds an --auto-review flag; when MCP executes automatically, it skips human interaction, and the output is marked with review_note for the caller to verify key figures

• Execution isolation: each run writes a temporary input file (deleted after use), and output is persisted to the template's output/

Quick start

pip install -e .
python -m workflow_mcp.server                 # stdio（默认）
python -m workflow_mcp.server --transport streamable-http

Prerequisites: the workflow-templates repository is on this machine at E:\NEMB\workflow-templates (can be overridden with the WORKFLOW_TEMPLATES_HOME environment variable); the template's own .env (e.g., OpenRouter API key) is loaded by run.py itself.

End-to-end test (will actually run the financial report analysis template once):

python examples/client_test.py    # stdio 协议方式
python examples/inproc_test.py    # 进程内客户端方式（CI / 沙箱环境更稳）

Integrating with WorkBuddy

Merge examples/workbuddy_mcp.json into mcpServers in C:\Users\33033\.workbuddy\mcp.json, then click "Trust" on workflow-mcp in the connector management page to enable it. After that, you can simply say "run the financial report analysis template for me" in a conversation.

Relationship with the trio

Roadmap

• v0.1: list / get / run three tools + --auto-review (current)

• v0.2: persist template outputs to a database (traceable results), concurrent execution queue, n8n template import

• v0.3: combine with digital-twin-mcp into a "device health report" template (data pipeline × brainpower pipeline)

License

MIT License